A precise deep learning approach for timber traceability along the forest value chain

Le résumé fourni par la source

The European Union Deforestation Regulation (EUDR) mandates traceability of timber that makes up wood products from its harvest site to the end product to ensure sustainable wood sourcing. This study proposes a cost-effective, image-based method for tracing logs using alphabetic codes printed onto logs at the harvest site. These codes are detected and interpreted through a two-stage system leveraging deep learning models. The detection stage employs YOLOv8 to locate tracking codes in images of log piles. It is trained and evaluated on a dataset of 125 images, achieving an F1-score of 0.811 on unseen images. The recognition stage, trained on 1,020 images, uses YOLOv8 models to detect individual characters and their positions within each code. On a set of unseen images, the interpretation stage is able to identify 92.8% of the individual logs despite the limited quality of the printer and degradation of the codes due to stem wetness. Analysis indicates that errors predominantly arise in the character detection step. Compared to existing traceability approaches, this method is more cost-effective than RFID tags and attains higher accuracy than image-based biomarker tracking methods.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.